An Research On Point Cloud Registration Algorithm Based On Kinect Device

At present, with the development of the computer vision, reconstruction techniques of 3D scene model has been one of the core techniques of computer vision.It attracts attention researchers and scholars. This paper mainly studies the reconstruction techniques of three-dimensional indoor scenes based on rotating device. Using rotating platform makes it rotate on a fixed axis and it gathers point cloud data from every angle. then it generates a complete 3D model of indoor scenes. Therefore, this paper gathers three-dimensional point cloud data from every angle and generates a complete 3D model of indoor scenes. It discussed some key techniques of point clouds registratiion. It mainly includes acquisition and filtering of point cloud, multi-viewpoint point cloud registration, integration and reconstruction.Acquisition and filtering of point cloud:this paper is based on the rotati ng device platform.It consists of a device sensor and an accurate control of r otating platform. With the help of a device sensor,it gathers depth information and RGB value. through coordinate transformation it converts the depth valu e to three-dimensional coordinate of point cloud data. Kinect acquires amount of information and it contains unstable vibration noise.This paper adopts an new bilateral filter of MQ. It expand from the 2D image processing into 3D point cloud data processing.Registration of multi-viewpoint point cloud data is an important step that it build a complete 3D of indoor scenes. It uses the characteristics of the rotating kinect platform to calculate coordinates transformation of multi-viewpoint collected point cloud data information. so the different space coordinates of the point cloud data unified into the same coordinate system, realize the point cloud data of coarse splicing. Currently widely using registration algorithm is the ICP algorithm. This paper make full use of the characteristics of rotating kinect platform to propose a new registration of multi-viewpoint point cloud data based on umbrella volume integral invariants algorithm. the curvature of umbrella is the invariant of rotation, translation and scale. It combines with strong identification and robustness of the volume integral invariants to extract the key points and calculate its feature descriptor. Combining coordinates of character description in the data set, it calculates corresponding points and estimate the coordinates transformation. So it implements registration of multi-viewpoint point cloud data.The integration and reconstruction of point cloud data is also important. the reconstruction of three-dimensional indoor scenes refers to registration of multi-viewpoint point cloud data. This paper use the poisson equation of curved surface 3D point cloud data fusion algorithm. It can reconstruct three-dimensional indoor scenes. This paper have multiple sets of experimental to gather point cloud and registration of indoor scene. The experimental result show that proposed new method of point cloud data registration is high precision. It is fast and visual result is satisfactory.